On Mon, Sep 29, 2003 at 07:55:42PM -0600, Evan McNabb wrote: > Because last week was so busy with UUG events we'll take this week off. > So there won't be a meeting on this Thursday (2nd). We'll be pushing it > to next Thursday (9th). > > The meeting is on the Linux Kernel.
I just took a week-long intensive class (cramming a semester's work of
lectures into 5 8-hour days) on the Linux 2.6 kernel here at work.
Gooooood stuff. :-) Andrew should have some very interesting facts
about the 2.6 kernel to point out. Personally, I am impressed with
all the cruft that was cut right out of some of the core structures.
There were a number of section where Linus & Friends said, ``This is a
relic of the ancient past. Let us be gone with it once and for all!''
<RIP>
The 2.6 kernel is, theoretically, not as efficient as the 2.4 kernel
was. That is, there are more opportunities for interruptions and
context switches, which guarantees that less time will be spent on raw
number crunching. But what really matters is that the user perceives
that the system is more responsive, and Linus cares more about the
workstation than the server.
Insanely fast processor speeds and good hardware support for context
saving and restoring help mitigate the overhead of the added
preemption code.
Oh, and have I mentioned the new hardware drivers? My USB webcam now
works with a vanilla 2.6 kernel. :-)
2.6 Trivia:
- The swap cache is no longer required for the page stealing process.
- The NUMA system maintains the concept of ``memory pressure'' on a
per-node basis and dynamically balances requests across nodes.
- There is no prioritization of interrupts (okay, so this is not new,
but it's still interesting).
- Major and minor numbers are no longer used internally to reference
devices (dev_id is now an opaque handle).
- ``Bottom half'' software interrupts were removed from the kernel.
- ``Task queues'' have been replaced with ``work queues.''
- No special block size is assumed for a device.
- getblk and bread were deprecated; ``bio'' services should be used
instead when writing device drivers.
- All block I/O is, by default, cached in the page cache (yes, this
implies that the buffer cache was removed in 2.6. Consolodation is
a Good Thing(tm)).
- We may not live long enough to see a ``3.0'' Linux kernel (my
personal opinion :-)
Mike
.__________________________________________________________________.
Michael A. Halcrow <[EMAIL PROTECTED]>
Security Engineer, IBM Linux Technology Center
GnuPG Fingerprint: 05B5 08A8 713A 64C1 D35D 2371 2D3C FDDA 3EB6 601D
If you are too busy to read, then you are too busy.
pgp00000.pgp
Description: PGP signature
____________________ BYU Unix Users Group http://uug.byu.edu/ ___________________________________________________________________ List Info: http://uug.byu.edu/cgi-bin/mailman/listinfo/uug-list
